Physiological Denoising Method for Unbiased Analysis of Biomedical Signals: Application on Heartbeat Dynamics

Background: Physiological systems show nonlinear deterministic behavior influenced by dynamical stochastic components, also known as physiological noise. Those components may bias deterministic system modeling and characterization. Objective: This study presents a model-free physiological denoising method for biomedical signals, such as heart rate variability (HRV) series, specifically focusing on the reduction of dynamical noise. Methods: The proposed method employs state-space reconstruction and time-reversing one-step forecasting, selecting optimal values within a neighborhood in the multidimensional space. The neighborhood size is determined as proportional to the physiological noise power. Synthetic time series analysis validate the correctness of the proposed method. Real HRV series from healthy subjects, patients with congestive heart failure, and those with atrial fibrillation were denoised, and unbiased complexity analysis was then performed. Physiological denoising performance was evaluated using root mean square error and median absolute deviation statistics. Results: Synthetic data analysis on canonical nonlinear maps demonstrated that the proposed method outperforms existing techniques in dynamical noise reduction. For HRV series, the proposed method effectively reduced physiological noise while preserving signal characteristics such as mean. While Sample Entropy analysis on original HRV series associated atrial fibrillation with the highest irregularity, unbiased analysis on denoised series revealed that healthy individuals actually exhibit the highest cardiac complexity. Conclusion: The proposed method effectively performs physiological denoising in biomedical signals, providing a reliable tool for unbiased analyses. This method enhances the understanding of underlying physiological dynamics that are intrinsically influenced by stochastic components.